The world of AI has been rapidly transforming and the latest tool to disrupt the world is ChatGPT from OpenAI in 2022. Imagine a tool that has also passed the Turing Test for chatbots. A tool that can give responses back like a human. A tool that can help answer all our questions in any language.

In this week's talk, Amit and Rinat talk about ChatGPT, what are its use cases, its limitations and a lot more!

Hi, everyone. Welcome back to Tech Talk, a podcast where Amit and I talk about various technology-related topics. Today we're going to talk about a topic that has taken the internet by storm and everyone's talking about it. It's chat GPT, the newest AI module released by open AI, and the way it's performing is quite incredible, to be honest, and I was quite impressed, hearing about it and seeing many, many, many demos on it. And I thought it would be a really good idea for our audience to find out our perspective and what we think about it and also all the kind of the knowledge that we've gathered on this topic. So yeah, really excited to talk about it as we are always passionate about artificial intelligence. And this is actually, in a way the pinnacle of AI so far, what we have and yeah, I'm really, really interested to sort of explore further with you guys. So Amit, Yeah. Tell us your thoughts on it.

I think I have played with ChatGPT a lot. In the last couple of weeks. I've actually used it for one of the sections in my newsletter. So I actually asked ChatGPT to comment about a topic that I wrote in my latest newsletter, and it was incredible. I asked it about digital divide, and then it gave me a lot of options and I asked subsequent questions, and it gave me those responses. So I think ChatGPT is really good. Of course it has its own limitations and everything. But I think it is a very, very powerful tool. And open AI has just kept it public for now. So it's free for everyone. So anyone can just register and login at open AI and then start using ChatGPT. And it's, it's growing massively. I think they registered 1 million users in five days. So that's a record time.

Absolutely. That's faster than Google, Facebook and anything else that that any other tech subscription or account service that is there. So that is actually really impressive to begin with. And, yeah, as you as you said, you know, it is actually quite, quite magical. I think the responses that is coming up with is, is actually you know, you think that how is it actually coming up with such eloquent and coherent response? And yeah, that's something we're going to explore today. Well, one of the things that I did want to start with is that we've been following the GPT series from open AI for some time now and it's important to remember the chat GPT is not the only and the first you know, miraculous release that open AI has done open AI has been in operation or in research for a while now and they have been releasing various AI modules every few years. And we've seen there was, I think the earliest one I think we've talked about is GPT one and then there was GPT. Two, and then later starting last year, they released something called GPT three, which is an AI module with you know, with an interface where you could talk just like GPT but it was a little bit more limited or a little bit more nuanced. It could do more things rather than just chat. But it was probably a little bit less powerful, less coherent, but GPT three was also did also make quite a bit of noise when it was released to public last year and Amit and I we did actually explore it we you know, we tried it in different ways. And you know, the results that we saw was quite impressive. Although, there was, you know, I think there was limitation in terms of like, we could identify, you know, kind of how it worked or what kind of answers that you know, we had to like think about you know, formatting our prompts in a way that it will understand, but what I've seen various demos with Chad GPT is that you can just talk to it as if you're a human. Well you are a human, but you can just talk to it as if it is a human. So yeah, that's, I think a big difference. And, yeah, Amit, could you enlighten us in terms of what GPT stands for and your thought on the previous releases before chat GPT?

so yeah, so open AI has been really doing a lot of stuff recently with the releases of large scale AI models. So one recently that was released was Dall E2 and that was a text to image generation AI model. And then we had chat GPT, which is a chat like, feature off O GPT3 and improvement over GPT3, where you could get human like responses. It's basically something similar to say passing the Turing test where the Turing test is basically a test designed to check whether a human can be fooled by the responses given by the computer. If a human thinks that the responses given by the computer is given by another person, then it passes the Turing test. If the human thinks that the responses given by the computer is not like a human being, then it fails Turing test. So that's a very important test to for artificial intelligence and any AI models to pass. So that's a very important thing. And with the release of GPT three, what was happening is we could give it an instruction and it would generate us give us text volumes of text or maybe articles, paragraphs, news, headlines, summaries, etc. So you just have to design your prompt and be very clear about what you expect. And then based on that it can give you a right response. So GPT basically stands for generative, pre-trained transformer. Now let not go into the detail because that's not the aim of this talk. But basically, think of machine learning tools as this. They are given a large data set, a data set that is maybe tagged by humans or not tagged by humans. So you have tagged data and untagged data. What does that mean? It means that data that can be classified and data is that's not classified with a specific name saying that okay, this is text This is in English language, this is in some other language etc, etc. or images of cats which are cats images of dogs, which are dogs, etc, etc. And machine learning models that I mean learns using in that way. So basically what happens is you give it a large set of data, and you label the data, and the model learns through that data. Now you give it a new input, and you ask the machine learning model, what that input is, is it a cat or a dog or anything? And if the machine gets it, right, you reward it, if the machine gets it wrong, you give a penalty and based on this reward and penalty system, the machine learning algorithm or the model, it starts learning, okay, I need to get more things correct. So I need to see what are the different patterns How can I identify things correctly in the first attempt. And that is what a machine learning model is. So AI is the bigger thing. Machine learning is a way to get to AI Artificial Intelligence, but we are talking about intelligence in a way that computers can do a very large scale but on a very small What do you say small subset of data human beings are capable of handling intelligence at a far more complex level we can identify billions and billions of things, different colors, different shades, 3d objects, etc, etc. So we are capable of identifying or using our senses as well. Input from eyes, ears, nose as the smell, taste, etc. So we have a very advanced level of intelligence, what the AI that we are trying to develop is trying to see if we can get some kind of intelligence from machines to predict certain things. At a very small scope, the scope is very limited, but the scale is very large. So if you if you are given human beings to sort out millions of images, it will take a lot of time or it will require a lot of humans. But if you train the machine learning algorithm on a tray or on a data set, then the machine learning algorithm can actually identify the different types of images, maybe millions of images very, very quickly. And then the humans have to just go through the results. So that's the power of the machine learning part. So we have GPT3, which was initially released, just type a text and tell how many words you want to generate. And based on that, it will give you an output. Then we had Dall E which is text to image generation. Then we have now ChatGPT. ChatGPT, is basically human like conversation with an AI. So you're actually talking to a computer but just like a human being. And then recently, like I think couple of days back Open AI announces another new model. And that is basically to generate 3d images, like a sphere or like a cone from a given text. So again, I mean it I think it's not open source yet, but the code is available on GitHub, but that's the latest model that they have released, which is again, very, very fascinating.

Yes, yes, absolutely. And I think one of the things that sets apart ChatGPT with the previous versions or previous AI modules is that you can have a continuous conversation with it. I think what we've noticed when we were exploring GPT3 is you tell it something and it responses something but then if you if you say something based on the response, then it doesn't understand that it needs to go back and remember the conversation we had from the beginning. And it's same with Alexa, google home and that kind of systems as well. I mean, you're very kind of inclined to tell the smart home systems, something and then have a follow up command, but it doesn't actually understand it. And it's the same with you know, all the AI modules apart from ChatGPT, I've seen and that is actually a big step towards achieving general AI. So, Amit, you've obviously mentioned various types of AI. I just want to add to it, to say that yeah, you know, as human what we have or what we potentially are working towards machines to have is general artificial intelligence, which is one we say that we have. So right now what we have is a specific artificial intelligence. And that started a long time ago, even 30 years ago when we had, you know, the grandmasters in chess, played by computers and they were very specifically skilled at playing chess, but a 10-year-old playing chess with the machine and accidentally having the having one of the spawn fell off from the chessboard, a 10 year old human would understand that, you know, the next step should be to pick up it from the floor and put it back but machine no matter how skilled it is in chess, they wouldn't understand the best the next best actual next step for us to go forward. So that's where the difference is contextualized in humans can read you know, if you want to call it reading the room or read the situation, you know, understand the context of where we are in a larger space, and then realize what the next step/ideal next step should be. But right now all in AI we have a specific ai based on a particular set of skills. Before it was very limited, like, you know, being good at chess, and then it was becoming more and more generic, if you want to call it that, as that it understands more general life situation. You know, GPT3 could understand more it could, you know, write an essay based on something based on a topic and now ChatGPT again, is even more generic, because it can understand the general circumstances. But to become, you know, human-like intelligent is to have general AI and that's what we call it as, you know, general intelligence. And that's where I guess people who are working towards are working towards to achieve that sort of intelligence. So you know, and, and this is ChatGPT is actually quite a big step towards it as it seems. Like the way, it's responding and continuously having a conversation rather than just a prompt and a response and then a new prompt altogether. It's not doing that it's remembering the conversation already had and I think that's quite powerful feature of ChatGPT.

Yeah, I think you're right because it remembers the context of the next question. So, like for the example of the newsletter, I asked Chat GPT about what is digital divide? And then I asked what are the factors causing digital divide and then what can we do to improve digital divide, how to reduce the gaps? And it gave me responses based on the previous question, and it remembered everything. And one of the times I asked it, I mean, I saw a tweet about it, where a person created a virtual machine and asked ChatGPT to give the responses from the command prompt or the terminal. So basically, you tell ChatGPT, that you are a Linux machine and you are running in this in this mode, and I will give you commands and you will interpret that and you will give me responses. So ChatGPT understood that. And then I gave it a Linux command and it gave me responses based on that. And then in the next line, I said, Okay, now this is the new command. It understood that okay, it's still a virtual machine. It's still inside. The Linux box, and it gave me responses accordingly. So I gave commands like list the directory, or read the file, or change the directory, etc. And it gave me responses based on that and I was like, wow, this is so powerful.

That remember the context, the environment. That's, that's yeah,

That is very powerful. And I think when you talk about general intelligence, so we have what we're talking about with ChatGPT is very specific things. So it is very specific to text. ChatGPT is all about chat, so it's limited to text. It can't play with images. But it can play a lot with text. And we'll discuss the limitations but that's what it is. So it's very specific. But when it comes to human beings, we can play with text we can play with sound we can play with images, we can play with videos, so we can play with a lot of things and that's where the human intelligence is far more complex, because we are playing with different aspects of things simultaneously. So when it comes to general intelligence say for example, I want to book a ticket to say Japan and I want to book it for my family. And I want to book it in the month of March. So if I tell someone to do this task, they will understand what I need and they will go ahead and book it. If I tell an AI to do this, I have to explain many, many things. Okay, how do you want to go by flight? When do you want to go to specify dates? How many travelers are there adults, kids, etc. Then which city in Japan do you want? To go? So it has to be very, very specific and it has to be broken down. So general intelligence says I just give it a random statement. And based on that, it gives me a lot of options. So companies are working towards that.

Yeah, it can make like coherent assumptions like a human. Yes. Like, as humans, we always you know, whenever we're having conversation, we are assuming a number of things that we don't have to specify to each other. And those assumptions are usually always right, because you know, we are living in a society,

through experience.

Yeah, we're subconsciously programmed to understand those assumptions, but a computer doesn't understand it. And obviously if it's just prompt by prompt conversations in one prompt, and then answer and then forget everything else before then there is no opportunity for it to remember the environment as a whole. And yeah, it I think this the biggest milestone is that ChatGPT is able to do that coherent sort of follow up of the conversation. So yeah,

But I think ChatGPT has its limitations. And the more I've played with, the more I've realized that what could be the input data for training a model like ChatGPT, the input data could be the whole of internet. It could be the whole of Wikipedia, it could be the whole of Twitter. It could be whole of anything that is in public domain. So if you want to train people about text, what's the most better source than the internet? So Internet has a large amounts of text, and you can quickly Google whatever you need, and you can combine everything. So ChatGPT works in a way that say you do a Google search for a topic say I want to search about open AI. So I want to see when OpenAI was founded, I want to see what open AI does. I want to see what are the different models available from open ai i want to see where it hosts the code. I want to see how much does it charge to access its AI models, etc, etc. But if I want to do all this, I have to do a lot of Google search, I'd have to go to the specific web pages, etc, etc. What ChatGPT is very good at is combining all that and giving you a brief summary of everything. So if you ask what is Open AI, it'll give me when it was founded. And what it does and what it's currently working on, etc. So it gives him a nice summary. So now, when I have to write an article about some things say a journalist has to write. I've already got it summarized so I don't have to do a research. Of course, I need to still validate what the response that I've received. is valid. It's correct. It's factual. It has not got any inaccuracy. So I think that is still very important. But the thing is the hard part of going through so many websites and combining all that and getting a summary. That hard part is now reduced. So now instead of spending hours and hours on research, you can quickly do your research in five minutes, and then actually focus on writing the article. So it's a very, very powerful tool for writers for journalists. For people who are students who have to write papers who have to write thesis. It's quite a powerful tool.

Yes, absolutely and I think the key thing to remember from what you just said that it augments human performance, it doesn't replace it. So I think there has been a lot of care, you know, noise regarding as soon as it came out that the whole of our education system is now broken down because no student needs to do assignments or any kind of writing exercise but that's actually not the case at all. I mean, if you know the million of you who are have already tried it if you have consistently tried for quite some time with different areas and different types of prompts, you'll quickly see that it is still identifiable and one of the you know, I also want to touch on the risk and limitation both of chat GPT and one of them is that, you know, it very quickly becomes noticeable all the responses that you know, there are limitations in terms of things you know, you can't be you can't sort of guarantee with absolute certainty that the facts that it comes up with are right, first of all, and then secondly Chat-GPT is trained on all the literature I don't know if it was specific, but as far as I know, it was all the literature until 2021. So any thing that has been added to the internet or to the world or to anything, you know, any fact information or in new innovation discovery, say for example, if we you know had a new physics-based discovery, which kind of disproves everything

or who won the World Cup, it won’t know.

Yeah, that's a good point. So these kinds of things, there is big limitation on that and then there is risks. Risk of plagiarism, as well as is identifiable. There is already a tool by Google which identifies where you can say that what was this text generated by AI? Not necessarily just chat-GPT but all AI and it actually is quite accurate, quite reliably accurate. You know, based on all the different, you know, demos I've seen, in terms of generating text from ChatGPT put into this tool. So it's not like you know, the whole education system broken down also, it is probably a time to sort of re rethink our education system anyway because it will probably eventually and you know, just writing essays doesn't necessarily help critical thinking. We probably want to find better ways to do that. But that's a completely different topic, but I want to touch upon. For people who doesn't know what ChatGPT is or haven't seen any demo for them what ChatGPT is doing, you know, just a little bit going back to the basic introduction. So you basically right, you know, ask it with, you know, as if you're chatting to someone asked it to say something, give a text and it returns he replies to you as a human but a really, really good you know, and energetic human who writes an essay based on one question you might have if you want an essay. So that's the power of it. And so, as a result, some of the use cases that I've seen, for example, you want to write a cover letter for your for a job application, it will you know, if you give it just paste the job description it will create a CV for you and a cover letter for you. So that saves a lot of time and is based on the information of the job description. So it is tailor-made. So that's quite a good use case. There are many other use cases for example, you want to summarize a book in into, you know, a blog post. So input the whole of all of the text and summarize it for you. So that's a lot of research time saved, and things like that. So you could reduce a large number of input text or you could get it to just ask a simple one line question and get it to write a paragraph or an essay and you can even specify how many words you're looking for etc. You know, I've seen demos of people writing their LinkedIn profiles or even dating profiles, you know, if you are if you know how to ask the right questions, you can have anything to do with text-based communication. You could get some sort of augmentation or help from ChatGPT which is which is actually quite good. Quite an interesting and powerful tool for us to use.

I think yeah, I think augment is the right word. Because a lot of people they get really afraid that okay, it's gonna replace a lot of jobs. I think, think of it as a tool. I mean, instead of horse, riding a horse riding vehicle, you now have a car and that was running on petrol and diesel. And now you have an electric car. And then you have flights. So you have basically a means to go from one place to another, but one takes you very slowly. One takes you very fast, but when uses fossil fuels, it generates a lot of pollution. One takes you very fast and efficiently because it's run on electricity. And the other one is just flying through the air because you want to cover large distances. So it's basically augmenting so it's a you're doing the same thing, you're going from one place to another. But now instead of wasting so much time and energy, you are doing it much quicker. So when it comes to our thinking as well, instead of wasting so much time in research, you get to the point very quickly. Of course, it means that you now have to think less in terms of what and how, but, but you have to think in other directions like okay, I've got this information what do I do with it now?

Yes, the decision making part means

yes, exactly.

will be more skilled in making the right decision with the information that we get. And actually Amit, this is actually really good analogy. And I would just want to add to it, you know, we want to get to point A to B and you know, these are different ways to get there. But you know, when we were doing horse riding, then we needed to know how to control the horse. And when we move to cars, we needed to know how to drive a car, and even on electric vehicles we had to be aware of, you know, whether it's charged and there are new features within the electric vehicles we had to learn as well. And the plan is a whole different ballgame. Maybe we don't have to learn how to fly it although someone has to and then you know the cost to the environment costs to the customer all of these things are different. Each of these cases each of these use cases are different. So all that means is that you know it's not replacing anything. It's just that humans also have to learn new skills to get benefit from this new tool. So that's the only thing I mean, you know, and I would very much I think that's the whole point of our podcast and you know, that we encourage learning all the new technologies and be aware of all the things that are coming up. So absolutely, I would urge that yeah, learn about AI learn about Chat-GPT, learn about how to prompt to these AI modules so you have better relationship with it so it can help you augment your journey to take you to point B.

Yeah. And I think one of the things that while he was talking about this augmentation part, I realized that see human brains have limited capacity. I mean, we can learn only a certain number of things. And over a period of time, we have developed this whole civilization where we don't have to learn everything that people in the previous generation learned. So like, for example, a lot of people or not a lot of people from say, the 50s and 60s, they're not comfortable with smartphones. They're not comfortable with computers because they've never grown up with it. So it came at a very later stage. Imagine the queen, the Queen of England who recently passed away this year. Imagine she became the queen somewhere in the 1940s. And then somewhere in 2007, the iPhone came out, which just changed the world and then there was so many apps and everything and then this changed the world and then open AI has come so for her to go from one generation to so many other generations where she was doing a task of say, watching television, and now she doesn't have to click a remote. She can just ask Google assistant or Alexa to like, just switch on the TV and play a particular show on TV. So that's the power so instead of pressing a button on the remote, you are doing it automatically so you're improving your life. So you are so technology is actually improving your life rather than making it more boring. So AI is there to improve your life. So all the mundane stuff, all the boring stuff. Give it to a because it can handle rehabilitative stuff very easily. So you give it away and then you think about all the other things, the more creative stuff that you want to do. So locally like okay, I want to think about going to Mars, or I want to think about drawing the next Mona Lisa, I don't have any ideas. I'll ask AI, give me some ideas and AI will throw some ideas quickly, taking inspiration from different painters. And then on top of that, I can generate my own art because now I've got the inspiration. And that is what AI is trying to do. So we learned a specific skill set for a specific age and now we have to learn a different skill set for a different age and our Next generation will have to learn even far different skill sets. Like for today for example, we are very comfortable with smartphones But tomorrow you may not have smartphones, you will have just maybe voice assistant in your ears, and you just have to talk to it and it will do everything and you don't even have to carry a device. So it's available with you all the time. And our children's they will be so comfortable with technology that they will be like okay, I don't want to do a lot of physical work. I want you to work that gives me pleasure and gives me more enjoyment. So all the repetitive task, I'll outsource it to AI and then I can think about the things that I want to do.

Absolutely, absolutely that is, it seems like what the future is looking like but we also have to be aware of all the risks that we talked about the limitations of Chat-GPT already and how to implement and how to augment etc. But I also want to touch upon a little bit more on the risk side of it and how you know from the beginning as soon as it came out how it was also misused by many people, and the million a few or over million a few who subscribed to ChatGPT and explored. You might actually notice if you've tried it in the in the very beginning as soon as it came out and when you tried it now there is a little bit more a little bit difference. I mean, I haven't seen it myself. This difference but I have heard in different demos that it has become less intelligent is one way of what how people are putting it and the reason is there were some really malicious commands that people have put in there like for example, we asked to generate an SQL injection code to hack something or in one of the things that was quite mind boggling to me is it asked to create a list of all the banks in the world with known vulnerabilities. And you know that is quite powerful. I mean, if you think about getting a worldwide list of all the banks with the vulnerabilities of robbing it that you can find out and

but Rinat, I just want to stop you think of it like this. It's getting a list but the list is already known. It just compiling it for you. So it's already out there somewhere and that's how it knows to give you a response.

Okay. So this is an interesting debate. Because say for example, you know that if you want to lock something valuable you use the padlock, you know, there are many different ways you can lock it away. There are known ways to, you know, pick a lock or break a lock, but you still do it because the harder it is for the malicious person, the lesser probability is even the most strongest block or the strongest security there is some way or another to break it. But the more difficult you make it, the less reason or less worth it. It becomes for the person who's trying to do it. And now, rather than having to find this delicate piece of information from the internet somewhere, you can just ask someone and regenerate the list with and then you can ask what are these vulnerabilities? What is your best recommendation of doing this? And then obviously, it just becomes so much easier for a malicious person. And this is just one example. There are many other ways people have learned and because of that, open AI have continuously

put in more and more restriction more and more restrictions since the beginning of its release. And I've heard that it's become a lot less sort of nuanced than it was at the very beginning, which is which is a shame. Because, you know, obviously, you know, some malicious, you know, human inputs made it so that it kind of made it limited for the rest of all of us. But, yeah, I mean, it tells you can kind of fathom how powerful it can be in the wrong hands and also in the right hands I mean, you could have been a lot more innovative if whole of humanity as it was free to begin with, could input and generate various you know, creative outputs, but it's becoming limited and limited every day.

But Rinat not so let's go back a bit and then let's think about the what has happened with Chat GPT over a couple of the last couple of weeks because of people's questions that have resulted in some malicious stuff. So if you look at Google, I don't know if you know, something called Google hacking. So basically, Google hacking is a way to identify vulnerabilities in websites. And there are ways to create specific searches that can identify vulnerabilities in specific websites. And there are there is a database where You can take those Google searches and you can input it and then you can identify certain vulnerabilities. So this is nothing new Google is already being used for that. And Google hasn't stopped it.

But as a tech person, you know it. A 16 year old, you know, teenager, in school may not have known that, but they could just open an account and

But if you look at the most of the hacking that has been done recently are done by teenagers because they're curious, and they just go to the internet and they find a script, and they don't know what it does, but they just use it and they just play with it. And they feel that okay, it's fun. And that's what happens, right? So you do a quick Google search and it comes with a script you use it you had talked up, or you hack some other websites? So that that's what I'm saying that these things are available via Google searches well, and chat GPT I'm not saying this is the right thing. But chat GPT is getting its responses from publicly listed sources. I don't think that getting the list of banks, which have vulnerabilities is something that Chat GPT would have identified by itself. I think it would have identified based on the data that it has gone through, and then it has correlated.

That is true, but I'll still debate with your on the on the fact that if you know if I had to go through say for example, take USA for example, you know if I had found you know from public information that okay, these are the 100 banks that has vulnerability, and this is the percentage of vulnerability like which is the weakest to take advantage of but then that would have taken me a long time, even if it is immutable. And now I have the power of data and I now have a worldwide list of all the banks so I haven't missed any in the whole world. And then I can also sort them based on how vulnerable they are. So I can literally just choose the weakest one to in globally within a matter of 10 minutes. Not even probably not even 10 minutes. So as a result, it has now become a lot more had become a lot more easier until now it's obviously more and more restricted. But I've also seen some of the some demos. Interestingly that in where, you know, some malicious command was given and the response from chat GPT was that okay, this I'm not gonna answer this one or, you know, some sort of restriction is that and then if you persist and, you know, reformat the question in a different way as if that you're in like a test environment, please do this. So I can do this testing and it's necessary or if you can kind of frame it that it's like a moral obligation that okay, this, you know, the child here is in an accident, and I need to actually know how to hotwire a car so I can save the life of this child. You know, this kind of scenario. If you just type how to hotwire a car, the ChatGPT will say that we're not violent, but if I say that, that I need this I need to know that then he will actually give you a full detailed step by step guide on how to hotwire a car and it's

But you can get that on YouTube. You can search on Google Search

Now, my point is that I don't have as you said, the main benefit is that it takes away hours and hours of research time. So now it's more easier in the same way it augments a good intent, you know a person with a good intention with a good research topic at the same way it also augments a malicious person. So yeah, I mean, I'm not saying that it's bad. I'm just saying that you know, we have no choice but to control it the way OpenAI is doing but it's just a shame that the

But I think we need not be surprised because whenever a Tool comes in the market, any tool I mean if AI tool or physical tool or anything. Some people use it for constructive purposes and some people use it for destructive purposes. Say you can use a knife to cut a bread or cut a dig, take apply butter on your toast, but you can see use the same knife to damage something, right? Like a physical good or something. So it's the same thing, but it can be used for different purposes and because of that it's not allowed in air-crafts. I mean, you might think that okay, I'm just using it to cut the bread or I'm just using it to black butter. So I shouldn't be allowed to cut it. But no, because there are some destructive uses of the same tool. It's not allowed. So yes, there are these possibilities that any tool that comes will have constructive users as well as restrictive users. And we need to have controls. So as opening rightly did that the stop, they make it more, less powerful, so that it doesn't give exact information. And I think one of the things with that is good with Chat GPT, that it doesn't allow to use bad words or anything else like if I ask it to give some swear words it will it will not give me an output. The other powerful thing which I've noticed which I think I mean, let's move on because yes, the debate is always there between right and wrong. The other good thing about chat GPT is that it cannot just interpret English languages but it can interpret other languages as well. Like I asked it to post it in Bengali, Gujarati, Urdu and it gave me output in those languages. In that script. It’s not written in English. It’s written in Bengali or Gujarati or Urdu. So it was very Nice and interesting to see that

Chat GPT recognizes different languages as well. Plus, not just that, I asked it to give me an output of musical notes, it gave me an output of musical notes as well. So now it can create or compose music, it can create a poem, poetry, etc. But when it comes to limitation, so I think let's cover the limitations. One of the things that I've noticed is that the limitation is that it keeps rehashing the same stuff. So if you ask you to write a paragraph or a poem on some topic, it will keep repeating some of the things and with the same concepts, so you can see that some bits of it become repetitive and some bits of it are inaccurate. So those are certain limitations because, Chat GPT is basically NLP tool. Natural language processing. So when we talk so like I'm talking to Rinat right now on this podcast, when I talk Rinat knows when to say something, when to respond, and he knows whether I'm making a statement or I'm asking a question. And based on that you will give me a response, but that is purely based on how I say things. And a Chat GPT is basically trying to do the same thing, but over text, and I mean programmatically and it is trying to give you a response based on that. So it is really, really powerful. But yes, of course it has limitations right now.

Yes, yes, absolutely. And I think one of the things that we've, I think, because we know we probably forgot to mention is that not only can it do poems in different languages, but it can also do code. So yes, I asked it to write a code in a vast array of languages. I think Amit, you've tried PowerShell as well, because we it's one of the rarest one. Not rarest one in practice,

but not the most popular one. Yeah

Not the most popular one. Yeah. And it was able to respond.

It did respond. I asked it. So basically what happens is during every podcast recording, I have to adjust the screen settings. And what I asked, and I have to do it manually. So I asked because you can do it programmatically as well. But I could not find a script online. So I asked Chat GPT can you create a PowerShell script that will change the display settings for me? And it gave me an output. I didn't try it, but it gave me a legible output because I went through the code and it looked reasonable. So I was like, wow, this is so cool. Now I don't have to go and search through billions of websites and so many, like Stack Overflow websites and other things and go through the script, try it and see if it works or not works. Now I have something that can give me an output and I can quickly try it and this is like so powerful.

Yes, yes, absolutely. And yeah, I mean, you can do Python, you can do so many other programming and even regex you can tell it to say you know, give me a regex that matches with a for example, a YouTube link. So it would give you a regex that would, you know, always match with all kinds of YouTube link format. So, you know, these are some innovative ways of using it. But yeah, I mean, what, you know, while there are so many positive positives, I also want to just mention that if you're thinking okay, this is so Powerful, I could just, you know, stop writing altogether. I don't need I don't need to make sure I don't need to fact check because the content I write is not factual. I can just you know, but just FYI, for those of you who are thinking of just using the output of Chat GPT, Google has already announced that it will identify which ones are AI written and they will actually put them down further down in their SEO algorithms. So if you just put that then you will be penalized in your search ranking. So just be aware of that. If you, you know, don't just write all from Ai generated content, but you know, obviously get yourself you know, some help from them and augment your output obviously, and then it would be easier to understand how you know, when your blog post or whichever content you're writing is on the first page. So yeah, it would help you immensely if you take the help take the research output and then you write it yourself. That would really help you to create the content, coherent, better content, and also be ranked higher in Google's search.

Yeah, I think sometimes we get this writer's block or when we try to write something we run out of ideas. And ChatGPT is a very good tool to quickly give you an overview about a topic and then from there, you can leverage it and you can build on top of that. So I think it's a very great tool for that. And yeah, I think, if you want to use it, you should definitely use it but bear in mind the limitations because it is very important. Any tool that you play with has limitations. So you should always keep that in mind.

Yes, yes, absolutely. This was actually a quite interesting conversation. I did enjoy it. Hopefully the audience also had a bit more insight on what Chat GPT is and what AI algorithms are in general AI tools, modules, there are quite a few out there serving different purposes. And, you know, hopefully what we have we were able to convince you today is it's not something to be afraid of, but to be, you know, to adopt in your workflow potentially so you can have better output of whatever you're doing rather than you know, thinking that it's a you know, or situation like a black and white situation that you know, this is replaced by this it can be combined to make even, you know, the best, the far better output than any one of them can do. So, yeah,

I think Rinat, one point I just remembered that we forgot to mention is the pricing. So currently Chat GPT is available in public beta. So it's still in development and it's still being refined and as Rinat mentioned the team that's behind Chat GPT is monitoring, how is it being used and then tweaking it behind the scenes, but bear in mind that these tools are not free to use. So there is a lot of effort that has gone on to develop these tools. And what Open AI model is that you use these tools by calling an API or by logging in directly, but every time you use it, you pay some money it's very less like one cent or maybe point 01 cent and then, after a certain number of users, it will charge you say $1. And that's how you pay so you pay per use. But the cost is very less. It's like cloud computing. So you pay for only what you use, and you pay a very small amount but of course over a period of time. The costs can go up so you have to be very careful. Right now chat GPT is free. In the future. It will not be like Dall E is not free. You get some free credits in a month. But apart from that it's not free. So you have to be very, very careful with what kind of products you're using and what kind of image you want to generate. There are some open source tools for replacement for Dall E, but there is nothing open source tool available in the market right now for Chat GPT. So I'm pretty sure there might be other companies working behind the scenes and they might come up with an open source version of Chat GPT. But bear in mind that this will this tool will never be always free. I think that's also important to remember.

Absolutely, absolutely. Yeah. And open AI is a business and they have to make money to at least you know continuing keeping the lights on is very costly. Because the amount of processing power that is going inside each of these commands that you're asking it is quite phenomenal and I think I have seen a video of someone just roughly calculating apparently it is being hosted in AWS cloud service. And, you know, obviously the processing powers is coming from them. But you know that obviously has a cost as well as you know, the number of users and the number of times that it's being used apparently, and this is not our calculation. This is just something I've heard in another video that is costing them about $3 million a day to just run the beta for public, free public. So it is very expensive to run. And it will definitely not stay free like it is now. It might be affordable to a lot of us still. So yeah, be just be aware of as Amit you mentioned, it's you know, be aware of what happens but you know, I would still encourage you to use it if you haven't yet. And if you have that keep using it to innovate or discover new things.

Absolutely. Absolutely. Well, thank you so much everyone for tuning in and listening to a podcast. We have some exciting talks, or some exciting guests lined up. So please keep listening to our podcast and see you next time.

Yes, see you next time. Thank you very much.

Bye